add optimization stage

README.md

@@ -50,4 +50,12 @@ At this stage, we get familiar with the concept of cross compilation, build and
 
 ### Porting Stage
 - [x] Alter the makefile for LoongArch cross compilation.
-- [x] Cross compile llama.cpp to LoongArch64.
+- [x] Cross compile llama.cpp to LoongArch64.
+
+### Optimization Stage
+Thanks to [the excellent work from Loongson team](https://github.com/ggerganov/llama.cpp/pull/6454), we have a great oppotunity to learn about SIMD acceleration with LoongArch LSX/LASX vector instruction set. Part of our work are based on them.
+- [x] Add LSX/LASX SIMD support for GGML_SIMD in `ggml.c`.
+
+### Benchmark Stage
+Benchmark goes along with optimization because we always want to know the exact improvement.
+- [ ] 

docs/tools_setup.md

@@ -1,21 +1,25 @@
 # Set up the cross compilation and emulation tools for LoongArch
 
-
-- Use the officially provided [build tools](https://github.com/loongson/build-tools).
-- Download the binaries of [cross toolchain](https://github.com/loongson/build-tools/releases/download/2023.08.08/CLFS-loongarch64-8.1-x86_64-cross-tools-gcc-glibc.tar.xz) and [QEMU linux-user](https://github.com/loongson/build-tools/releases/download/2023.08.08/qemu-loongarch64).
+- Use the officially provided [cross tools](http://www.loongnix.cn/zh/toolchain/GNU/).
+- Download the binaries of [cross toolchain](http://ftp.loongnix.cn/toolchain/gcc/release/loongarch/gcc8/loongson-gnu-toolchain-8.3-x86_64-loongarch64-linux-gnu-rc1.3.tar.xz) and [QEMU linux-user](https://github.com/loongson/build-tools/releases/download/2023.08.08/qemu-loongarch64).
 - For convenience, set the root dir of build tools and qemu as $LA_TOOLCHAIN and $LA_QEMU, respective. Add $LA_TOOLCHAIN/bin and $LA_QEMU/bin to $PATH.
 
 
 ## Basic Testing
 
 Test with C
 ```bash
-loongarch64-unknown-linux-gnu-gcc hello_loongarch.c -o hello_loongarch
-qemu-loongarch64 -L $LA_TOOLCHAIN/target/ -E LD_LIBRARY_PATH=$LA_TOOLCHAIN/target/lib64/:LD_LIBRARY_PATH  hello_loongarch
+loongarch64-linux-gnu-gcc hello_loongarch.c -o hello_loongarch
+qemu-loongarch64 -L $LA_TOOLCHAIN/loongarch64-linux-gnu/sysroot/usr/ -E LD_LIBRARY_PATH=$LA_TOOLCHAIN/loongarch64-linux-gnu/sysroot/usr/lib64/:LD_LIBRARY_PATH  hello_loongarch
 ```
 
 Test with C++
 ```bash
-loongarch64-unknown-linux-gnu-g++ hello_loongarch.cpp -o hello_loongarch
-qemu-loongarch64 -L $LA_TOOLCHAIN/target/ -E LD_LIBRARY_PATH=$LA_TOOLCHAIN/loongarch64-unknown-linux-gnu/lib/:LD_LIBRARY_PATH  hello_loongarch
+loongarch64-linux-gnu-g++ hello_loongarch.cpp -o hello_loongarch
+qemu-loongarch64 -L $LA_TOOLCHAIN/target/ -E LD_LIBRARY_PATH=$LA_TOOLCHAIN/loongarch64-linux-gnu/lib/:LD_LIBRARY_PATH  hello_loongarch
+```
+
+Test LASX support
+```bash
+loongarch64-linux-gnu-gcc test_lasx.c -o test_lasx -mlasx
 ```

la-llama.cpp/Makefile → llama.cpp-b2430/Makefile

@@ -95,8 +95,8 @@ CXX	:= riscv64-linux-gnu-g++
 endif
 
 ifdef LOONGARCH_CROSS_COMPILE
-CC  := loongarch64-unknown-linux-gnu-gcc
-CXX := loongarch64-unknown-linux-gnu-g++
+CC  := loongarch64-linux-gnu-gcc
+CXX := loongarch64-linux-gnu-g++
 endif
 
 #
@@ -287,8 +287,8 @@ ifdef RISCV
 	MK_CXXFLAGS += -march=rv64gcv -mabi=lp64d
 
 else ifdef LOONGARCH
-	MK_CFLAGS   += -march=loongarch64 -mabi=lp64d
-	MK_CXXFLAGS += -march=loongarch64 -mabi=lp64d
+	MK_CFLAGS   += -march=loongarch64 -mabi=lp64d -mlasx
+	MK_CXXFLAGS += -march=loongarch64 -mabi=lp64d -mlasx
 
 else
 ifeq ($(UNAME_M),$(filter $(UNAME_M),x86_64 i686 amd64))